from transformers import WhisperProcessor, WhisperForConditionalGeneration
import torch

processor = WhisperProcessor.from_pretrained("IonGrozea/whisper-base_ro-80mel")
model = WhisperForConditionalGeneration.from_pretrained("IonGrozea/whisper-base_ro-80mel")

inputs = processor(audio_array, sampling_rate=16000, return_tensors="pt")
with torch.no_grad():
    ids = model.generate(inputs.input_features, language="romanian",
                         task="transcribe", num_beams=5)
print(processor.batch_decode(ids, skip_special_tokens=True)[0])

This card was auto-generated at upload time because the source folder had no README.md (e.g. a raw checkpoint). For the full training-config card, upload from the final_best/ directory instead.